Short-form video (Two Minute Papers) where Dr. Zsolnai reviews a recent simulation paper on collision detection and contact-rich physics. He contrasts an industry-standard simulator that fails to screw in a bolt with a new method that succeeds, then shows more complex demos (an "expanding Lockbox" with tens-to-hundreds-of-thousands of contacts, piston-on-disc, chain-and-sprocket) to illustrate robustness under friction and complex geometry. The new technique allows time steps ~100× larger but incurs ~100× greater computation per step compared to prior fast-but-incorrect methods; the presenter frames it as "the fastest simulator that actually works." The episode includes a sponsor segment for Weights & Biases (tools for tracking deep-learning experiments; free for individuals, academics, and open-source projects).
Questions this video answers
The "expanding Lockbox" experiment (locking pins retract, bottom falls).
Approximately 100× larger time step size.
Resulting computation time per step is still at least ~100× slower than previous (but incorrect) methods.
Individuals, academics, and open-source projects.
Chapters
Dear fellow scholars, this is two minute.
Papers with Dr. Zsolnai. I have to say, we haven't had a simulation paper in a while, so today's episode is going to be my way of medicating myself. You are more than welcome to watch the process. And this paper is about performing collision detection. You see, when we write a simple space game, detecting whether a collision has happened or not is is mostly a trivial endeavor. However. Now instead, let's look at the kind of simulation complexity that you are expecting from a Two Minute Papers video. First, let's try to screw this bolt in using an industry standard simulation system. And it is stuck. Hmm, why? Because here we would need to simulate in detail not only whether two things collide, they collide all the time, but we need to check for and simulate friction too. Let's see what this new simulation method does with the same scene. And oh yes, this one isn't screwing with us and does the job perfectly. Excellent. However, this was not nearly the most complex thing it can do. Let's try some crazy geometry with crazy movements and tons of friction. There we go. This one will do. Welcome to the expanding Lockbox experiment. So what is this? Look, as we turn the key, the locking pins retract and the bottom is now allowed to fall. This scene contains tens to hundreds of thousands of contacts and yet it still works perfectly. Beautiful. I love this one. Because with this simulation, we can test intricate mechanisms for robotics and more before committing to manufacturing anything. And unlike with previous methods, we don't need to worry whether the simulation is correct or not. And we can be sure that if we 3D print this, it will behave exactly this way. So good. Also, here come some of my favorite experiments from the paper. For instance, it can also simulate a piston attached to a rotating disc. Smooth motion on one wheel leading to intermittent motion on the other one. And if you feel the urge to build a virtual bike, do not worry for a second because your chain and sprocket mechanisms will work exactly as you expect to. Loving it. Now, interestingly, look here. The time step size used with the new technique is a hundred times bigger, which is great. We can advance the time in bigger pieces when computing the simulation. That is good news indeed. However, every time we do so, we still have to compute a great deal more. The resulting computation time is still at least a hundred times slower than previous methods. However, those methods don't count, at least not on these scenes, because they have produced incorrect results. Look at it some other way. This is the fastest simulator that actually works. Still, it is not that slow. The one with the intermittent motion takes less than a second per time step, which likely means a few seconds per frame, while the bot scrolling scene is likely in the minutes per frame domain. Very impressive. And if you're a seasoned fellow scholar, you know what's coming. This is where we invoke the first law of papers which says that research is a process. Do not look at where we are, look at where we will be. Two more papers down the line and two more papers down the line. I am sure even the more complex simulations will be done in a matter of seconds. What a time to be alive.
This episode has been supported by Weights and Biases.
In this post they show you how.
To use their tool to visualize confusion.
Matrices and find out who where your.
Neural network made mistakes and what exactly those mistakes were. Weights and Biases provides tools to track your experiments in your deep learning projects. Their system is designed to save you a ton of time and money and it is actively used in projects at prestigious labs such as OpenAI, Toyota Research, GitHub, and more. And the best part is that Weights and Biases is free for all individuals, academics and open source projects. It really is as good as it gets. Make sure to visit them through wnb. Com papers or just click the link in the video description and you can.
Get a free demo today.
Our thanks to Weights and Biases for their long standing support and for helping.
Us make better videos for you. Thanks for watching and for your generous support and I'll see you next time.
